RBF neural network based on dynamic clustering for face recognition

The performance of RBF neural network is deeply affected by the neuron number, centers and width factor in the hidden layer, which belonging to the area of the clustering analysis. In the paper, a new dynamic clustering method, which is a bottom-up searching algorithm, to find the hidden neurons center of RBF neural network is proposed. In the first phase, it is a dynamic clustering process, and in the second phase, the prior result is partition by real rough set theory. According the lower and upper approximation clusters, the centers position and the width factor are confirmed respectively. The face recognition based on the RBF neural network experiments is simulated. The experiment results show that the proposed method is valid and effective. The accuracy of the RBF neural networks is increased and algorithm is robust.